Add timeout option for webhook notifier.

Author: stevesg

config/notifiers.go

@@ -535,6 +535,9 @@ type WebhookConfig struct {
 	// Alerts exceeding this threshold will be truncated. Setting this to 0
 	// allows an unlimited number of alerts.
 	MaxAlerts uint64 `yaml:"max_alerts" json:"max_alerts"`
+
+	// Timeout is the maximum time allowed to invoke the webhook.
+	Timeout *time.Duration `yaml:"timeout" json:"timeout"`
 }
 
 // UnmarshalYAML implements the yaml.Unmarshaler interface.

notify/webhook/webhook.go

@@ -17,6 +17,7 @@ import (
 	"bytes"
 	"context"
 	"encoding/json"
+	"errors"
 	"fmt"
 	"log/slog"
 	"net/http"
@@ -112,8 +113,17 @@ func (n *Notifier) Notify(ctx context.Context, alerts ...*types.Alert) (bool, er
 		url = strings.TrimSpace(string(content))
 	}
 
+	if n.conf.Timeout != nil {
+		postCtx, cancel := context.WithTimeoutCause(ctx, *n.conf.Timeout, fmt.Errorf("configured webhook timeout (%s) reached", *n.conf.Timeout))
+		defer cancel()
+		ctx = postCtx
+	}
+
 	resp, err := notify.PostJSON(ctx, n.client, url, &buf)
 	if err != nil {
+		if errors.Is(err, context.DeadlineExceeded) && ctx.Err() != nil {
+			err = context.Cause(ctx)
+		}
 		return true, notify.RedactURL(err)
 	}
 	defer notify.Drain(resp)

test/with_api_v2/acceptance/send_test.go

@@ -464,3 +464,52 @@ receivers:
 
 	t.Log(co.Check())
 }
+
+func TestWebhookTimeout(t *testing.T) {
+	t.Parallel()
+
+	// This integration test uses an extended group_interval to check that
+	// the webhook level timeout has the desired effect, and that notification
+	// sending is retried in this case.
+	conf := `
+route:
+  receiver: "default"
+  group_by: [alertname]
+  group_wait:      1s
+  group_interval:  1m
+  repeat_interval: 1m
+
+receivers:
+- name: "default"
+  webhook_configs:
+  - url: 'http://%s'
+    timeout: 500ms
+`
+
+	at := NewAcceptanceTest(t, &AcceptanceOpts{
+		Tolerance: 150 * time.Millisecond,
+	})
+
+	co := at.Collector("webhook")
+	wh := NewWebhook(t, co)
+
+	wh.Func = func(ts float64) bool {
+		// Make some webhook requests slow enough to hit the webhook
+		// timeout, but not so slow as to hit the dispatcher timeout.
+		if ts < 3 {
+			time.Sleep(time.Second)
+			return true
+		}
+		return false
+	}
+
+	am := at.AlertmanagerCluster(fmt.Sprintf(conf, wh.Address()), 1)
+
+	am.Push(At(1), Alert("alertname", "test1"))
+
+	co.Want(Between(3, 4), Alert("alertname", "test1").Active(1))
+
+	at.Run()
+
+	t.Log(co.Check())
+}